AI-powered, conversational data modeling platform. Upload a spreadsheet, chat with an AI assistant, and walk away with a validated ML model running behind a live API and interactive dashboard. No code required.
Built for business analysts who know their data but don't write code.
- Upload a CSV -- AutoModeler instantly profiles it (row counts, types, patterns, anomalies)
- Explore via natural language -- "Which products are trending up?" returns charts and stats
- Shape features -- say "suggest features" or "apply transformations" to engineer features through conversation
- Model -- recommends and trains appropriate algorithms, shows plain-English comparisons
- Validate -- cross-validation, confusion matrices, feature importance, per-row explanations, and segment-level performance breakdowns
- Deploy -- one click to a live prediction API endpoint + shareable dashboard
| Layer | Technology |
|---|---|
| Backend | Python 3.12+, FastAPI, SQLModel (SQLite) |
| ML | scikit-learn, pandas, numpy |
| LLM | Anthropic SDK (Claude) for chat orchestration |
| Frontend | Next.js 15, React 19, TypeScript |
| UI | Shadcn/UI (Nova), Tailwind CSS, Hugeicons |
| Charts | Recharts |
| State | Zustand |
| Package Mgmt | uv (backend), npm (frontend) |
src/
├── backend/
│ ├── api/ # FastAPI route handlers (chat, data, features, models, validation, deploy)
│ ├── core/ # Business logic (analyzer, feature_engine, trainer, validator, explainer, deployer)
│ ├── chat/ # LLM orchestration (prompts, narration, state machine)
│ ├── models/ # SQLModel database models
│ ├── db.py # Database setup
│ └── main.py # App entry point
└── frontend/
├── app/ # Next.js App Router pages
│ ├── project/[id]/ # Workspace (chat + data + features + models + validation + deploy)
│ └── predict/[id]/ # Public prediction dashboard
├── components/ # React components (chat, data, features, models, deploy, ui)
└── lib/ # API client, Zustand stores, types
- Python 3.12+
- Node.js 18+
- uv package manager
- Anthropic OAuth token (via
claude auth loginorANTHROPIC_AUTH_TOKENenv var)
cd src/backend
uv venv
uv sync
ANTHROPIC_AUTH_TOKEN=sk-ant-oat01-... uv run uvicorn main:app --reload --port 8000cd src/frontend
npm install
npm run devThe frontend runs on http://localhost:3000 and expects the backend at http://localhost:8000.
# Backend (1557 tests)
cd src/backend
uv run pytest
# Frontend (680 tests)
cd src/frontend
npm test| Endpoint | Purpose |
|---|---|
POST /api/chat/{project_id} |
Streamed chat (SSE) |
POST /api/data/upload |
CSV upload + auto-profiling |
GET /api/data/{id}/profile |
Data quality report |
GET /api/data/{id}/query |
Natural language data queries |
GET /api/features/{id}/suggestions |
AI feature suggestions |
POST /api/features/{id}/apply |
Apply transformations |
POST /api/models/{id}/train |
Train models |
GET /api/models/{id}/compare |
Compare trained models |
GET /api/models/{id}/segment-performance |
Per-segment R² / accuracy breakdown |
GET /api/validate/{id}/explain |
Feature importance (SHAP-lite) |
GET /api/models/{id}/report |
Download PDF model report |
POST /api/deploy/{id} |
Deploy model as API |
POST /api/predict/{id} |
Make predictions |
POST /api/predict/{id}/batch |
Batch predictions (CSV) |
The chat interface understands natural language for every step of the workflow. Key chat-triggered features:
| Say something like... | What happens |
|---|---|
| "suggest features" / "recommend transformations" | Inline FeatureSuggestCard with transform badges + one-click Apply All |
| "apply the feature suggestions" | Applies all transformations, emits FeatureSuggestCard confirmation |
| "how does my model perform by region?" | Inline SegmentPerformanceCard showing per-group R² or accuracy with best/worst labels |
| "model accuracy by product" | Same segment breakdown, auto-detected grouping column |
| "generate a report" / "pdf report" | Inline ReportReadyCard with direct download link |
AutoModeler is built on the code-evolve
framework -- a self-evolving project builder. An AI agent reads vision.md and
spec.md, then autonomously implements features session after session.
# Run an evolution session (uses OAuth from `claude auth login` or env var)
CLAUDE_CODE_OAUTH_TOKEN=sk-ant-oat01-... ./scripts/evolve.shSee spec.md for the full feature checklist and JOURNAL.md for the build log.
Phases 1-8 in progress (foundation through advanced conversational features). 2,237 tests passing (1,557 backend + 680 frontend).
Recently shipped:
- Model performance by segment — ask "how does my model perform by region?" for a per-group breakdown with gap analysis and retraining recommendations
- Chat-driven feature engineering — the full Upload → Explore → Shape → Train → Deploy workflow is now 100% conversational
- Chat-triggered PDF reports — "generate a report" delivers a shareable PDF inline in chat
Phase 9 (onboarding, project management, export, responsive design) is next.
MIT